IMX Talks - From first-principles simulations to autonomous discovery: building a computational ecosystem for accelerated materials design

First-principles electronic-structure simulations have become essential tools for understanding and discovering materials, providing predictive insight that complements experiments. In this talk, I present highlights from the research of my group, where we develop computational methods and open infrastructures enabling materials discovery and providing microscopic interpretation of experiments.

I first present the Materials Cloud MC3D database [1], a curated starting point for exploring materials space with relaxed geometries and, soon, relevant computed properties (electronic structures, projected densities of states, and descriptors for material classes including electrides). To accelerate simulations, a central part of our research focuses on reduced-order Hamiltonians based on maximally localized Wannier functions, enabling efficient evaluation of electronic and transport properties while retaining first-principles accuracy. I discuss our efforts to automate Wannierization [2], and their application to create a comprehensive database of Fermi surfaces and to predict targeted properties such as superconductivity [3], thermoelectrics, and nonlinear Hall effects.

A key enabler of this work is our open-science software stack built around three pillars: the workflow engine AiiDA, the open data-sharing platform Materials Cloud, and the browser-based environment AiiDAlab. These infrastructures allow advanced simulations (such as phonons, spectroscopies, and Wannier-function workflows) to be executed through automated workflows and intuitive interfaces. These developments are underpinned by our verification and validation efforts [4], providing a solid foundation for both discovery workflows and machine-learning models. Finally, I briefly outline examples of ongoing work toward autonomous platforms, integrating artificial intelligence and Bayesian optimization with both simulations and experiments to accelerate materials design [5], and conclude with an overview of educational tools developed with EPFL colleagues to foster computational thinking in materials science [6].

[1]      S. P. Huber et al., Digital Discovery, Advance Article (2026)
[2]      J. Qiao et al., npj Comput. Mater. 9, 208 (2023) ; Y. Jiang et al., npj Comput. Mater. 11, 353 (2025)
[3]      M. Bercx et al., PRX Energy 4, 033012 (2025)
[4]      E. Bosoni et al., Nat. Rev. Phys. 6, 45 (2024)
[5]      P. Kraus et al., J. Mater. Chem. A 12, 10773 (2024); M. Vogler et al., Adv. Energy Mater. 14, 2403263 (2024); M. Vogler et al., Matter 6, 2647 (2023)
[6]      D. Du et al., Comput. Phys. Commun. 282, 108546 (2023); D. Du et al., Comput. Phys. Commun. 305, 109353 (2024); A. Goscinski et al., arXiv:2507.05734 (2025)

Bio: Dr. Giovanni Pizzi is group leader of the Materials Software and Data group at the Paul Scherrer Institute (PSI), Switzerland. He obtained his PhD in Physics from Scuola Normale Superiore in Pisa and then worked at EPFL, first as a postdoctoral researcher and later as a permanent scientist, before joining PSI in 2022. His research focuses on computational materials science, including high-throughput materials discovery, Wannier-function methodologies, verification of density-functional theory implementations, and the design of autonomous platforms for materials discovery and characterization.

He is the author of one patent and over 65 peer-reviewed publications. He leads the development of open-science infrastructures for computational materials science (AiiDA, Materials Cloud, AiiDAlab) and is a co-author of the Wannier90 code. He serves on the editorial boards of Computational Materials Science and Scientific Data. His distinctions include the 2010 "G. F. Bassani" Prize, the 2020 "R. & R. Haenny" Prize, the 2022 PRACE HPC Excellence Award (jointly), and the 2024 PSI Diversity Award (finalist).